Contributions to the mechanisms and simulation of mineral processing operations

Abstract
ENGLISH ABSTRACT: This collection of 101 published papers is based on 19 years of research, mainly in South Africa and recently also in Australia. The main thrust of this work was the simulation of mineral processing operations where conventional modelling was inadequate owing to their ill-defIned nature. These papers presented some of the first applications of knowledge based systems and neural networks to mineral processing problems. A new methodology was proposed for modelling ill-defmed kinetic processes by relating rates to process conditions via non-parametric methods. The kinetic and equilibrium parameters could be related to adjustment factors if reference conditions changed. These methods were applied to batch and continuous flotation, leaching and adsorption systems as well as pyrometallurgical processes. In the case of continuous process data intrinsic reaction rates could be back -calculated and then related to process conditions via a neural net. It was explained how the configuration of mineral processing circuits could be optimised by a two-stage linear programming method where the constraints were determined interactively by a knowledge based system or neural network. In the case of non-linear system constraints neural nets were shown to detect and locate gross errors efficiently in material balancing problems. Several novel hybrid neural net architectures were proposed which allowed the integration of parametric fundamental knowledge with non-parametric heuristic knowledge. Furthermore, it was shown how neural nets could be used to extract knowledge from historical process data records. Pioneering work was conducted on the application of textural image analysis to flotation froth surfaces and ore particles on a conveyor belt. It was possible to relate image features to metallurgical performance via neural net architectures. Different types of topological maps were used to classify froth types, to track the profile of changes in process and flotation conditions, and to relate froth class to concentrate grade and recovery. Perturbations in the comminution circuit were reflected in perturbations in the froth image features and metallurgical performance. It was also shown that the spray angle of a hydrocyc1one as determined by image analysis could be used as input to a soft sensor for predicting size distribution. A model was developed to combine transport phenomena in the froth phase with fluid flow behaviour in order to optimise mechanical cells and flotation columns. Rate constants for the various sub-processes were determined by fitting models to concentration profiles measured at industrial plants. It was shown that the behaviour of the froth phase during the flotation of sulphides, gold and uranium was affected by galvanic interaction between metallic iron and the individual mineral species. A conceptual model was developed for the flotation behaviour of free gold in the presence of refractory sulphides for different oxidative conditions and different sequences of reagent addition. The use of diagnostic leaching to evaluate the efficiency of mineral processing operations was explained in detail. A first attempt was made to relate diagnostic leaching data to mineral liberation via semi-empirical equations and neural nets. It was explained how the liberation patterns of different types of gold ore could be distinguished by using topological maps. The mechanisms of selected sundry processes such as electrokinetic solid-liquid separation, flow splitting, induced aeration and jet reactors were also investigated.
AFRIKAANSE OPSOMMING: Hierdie versameling van 101 publikasies is gebaseer op navorsmg oor 19 jaar, hoofsaaklik in Suid-Afrika en onlangs ook in Australie. Die fokuspunt van hierdie werk is die simulasie van mineraal prosesserings operasies waar konvensionale modellering ontoereikend is vanwee swak gedefmieerdheid. Hierdie publikasies toon van die heel eerste toepassings van kennis gebaseerde stelsels en neurale netwerke op mineraal prosesserings probleme. 'n Nuwe metodiek is voorgestel vir die modellering van swak gedefinieerde kinetiese prosesse deur die reaksie tempo's te verbind met proses toestande via nie-parametriese tegnieke. Die kinetiese en ekwilibrium parameters is in verb and gebring met verstellingsfaktore indien die verwysingstoestande verander het. Hierdie metodes is toegepas op enkellading en kontinue flottasie, loging en adsorpsie stelsels, asook op pirometallurgiese prosesse. In die geval van kontinue proses data kon die intrinsieke reaksie snelhede terugbereken word en in verb and gebring word met proses toestande via 'n neurale net. 'n Twee stadium lineere programmerings metode is voorgestel vir die optimale ontwerp van mineraal prosesserings aanlegte deur die beperkings interaktief te bepaal met 'n kennis gebaseerde stelsel of neurale netwerk. In die geval van nie-lineere stelsel beperkings is getoon dat neurale nette effektief buitengewone foute kon uitken en opspoor in massabalans probleme. Verskeie nuwe hibriede neurale net argitekture is voorgestel om parametriese fundamentele kennis te integreer met nie-parametriese heuristiese kennis. Verder is getoon hoe neurale nette gebruik kan word om kennis te ekstraheer uit historiese rekords van proses data. Pioniers werk is gedoen om tekstuur gebaseerde beeldanalise tegnieke toe te pas op flottasie skuim oppervlaktes en erts partikels op 'n vervoerband. Dit is moontlik om beeld kenmerke in verb and te stel met metallurgiese doeltreffendheid via neurale netwerk argitekture. Verskeie topologiese afbeeldings is gebruik om skulln tipes te klassifiseer, om die profiele van veranderings in proses en flottasie toestande te volg, asook om die skuim klas in verb and te bring met herwinning en konsentraat graad. Versteurings in die groottereduksie aanleg is afgebeeld deur ooreenkomstige versteurings in die skuim kenmerke en metallurgiese gedrag. Daar is ook getoon hoe die sproei hoek van 'n hidrosikloon soos bepaal deur beeldanalise, gebruik kan word as inset tot 'n sagte sensor om partikelgrootte verspreiding te voorspel. 'n Model is ontwikkel om oordragsverskynsels in die skuimfase te kombineer met vloeipatrone om sodoende meganiese flottasie selle en kolomme te optimeer. Snelheidskonstantes vir die verskeie sub-prose sse is bepaal deur modelle te pas op konsentrasie profiele soos gemeet op industriele aanlegte. Daar is getoon dat die gedrag van die skuim fase in die flottasie van sulfiede, goud en uraan bemvloed word deur galvaniese interaksie tussen yster metaal en die onderskeie mineraal spesies. 'n Konseptuele model is voorgestel vir die flottasie gedrag van vrye goud in die teenwoordigheid van weerbarstige sulfiedes onder verskillende toestande van oksidasie en volgorde van reagens byvoeging. 'n Gedetailleerde verduideliking is gegee vir die aanwending van diagnostiese loging om die doeltreffendheid van mineraal prosesserings operasies te evalueer. Diagnostiese loging data is vir die eerste keer in verband gebring met mineraal vrystelling deur semiempiriese vergelykings en neurale nette. Daar is verduidelik hoe die vrystellingspatrone van verskeie goud erts tipes onderskei kan word deur 'n topologiese afbeelding. Die werking van verskeie prosesse soos elektrokinetiese vloeistoflvastestof skeiding, vloei verdeling, geinduseerde belugting en spuit reaktore is ook ondersoek.
Description
Thesis (Ph.D.) -- University of Stellenbosch, 1999.
Keywords
Ore-dressing, Neural networks (Computer science), Flotation, Fluid mechanics, Leaching, Computer vision, Dissertations -- Chemical engineering
Citation